►
From YouTube: GitOps Guide to the Galaxy (E42) | Meet Robusta.dev
Description
What is Robusta and why do they advise against CPU limits on Kubernetes? https://home.robusta.dev/blog/stop-using-cpu-limits/
Want to learn more?
https://home.robusta.dev
Try Red Hat OpenShift: https://red.ht/3sMaW7O
Follow us:
Hilliary Lipsig https://twitter.com/Int2Caffeinate
https://twitter.com/aantn
OpenShift https://twitter.com/openshift
Subscribe to Red Hat's YouTube channel: https://www.youtube.com/redhat/?sub_confirmation=1
What is GitOps Guide to the Galaxy?
Every other Thursday at 3pm ET host Hilliary Lipsig and guests dig in to discuss everything in the GitOps universe, from end-to-end CICD pipelines to creating Git workflows.
Learn GitOps https://red.ht/3saqlzS
What is GitOps? https://www.redhat.com/en/topics/devops/what-is-gitops?sc_cid=7013a000002qG1BAAU
About OpenShift:
Red Hat OpenShift is an open source container application platform based on the Kubernetes container orchestrator for enterprise application development and deployment.
OpenShift on redhat.com https://red.ht/3LGCIeD
OpenShift on GitHub https://github.com/openshift/
#RedHat #GitOps #Kubernetes
A
B
B
So
for
those
who
who
might
not
be
old
enough
or
be
aware
of
Douglas
Adams
Hitchhiker's
Guide
to
the
Galaxy,
you
can
see
where
we
kind
of
got
the
name
from
the
number
42
is
the
answer
to
life,
the
universe
and
everything
we
still
do
not
know
the
question,
but
nevertheless,
today's
today
is
the
absolute
best
episode
a
couple
announcements
before
I
introduce
our
very
exciting
guest.
So
actually
it's
just
one.
B
You
will
be
able
to
see
me
talking
about
Helm
and
kubernetes
operators
at
Devcon,
Us
in
Boston,
Dev,
conf
U.S
in
Boston
on
August,
I,
think
I'm
speaking
on
the
17th,
but
anyway
find
me
there.
If
you're
going
to
be
there
in
person,
say
hi.
That
is
a
really
great
conference.
Lots
of
people
will
be
there.
Other
people
from
my
team
and
organization
will
be
there
as
well.
B
So
it's
a
great
opportunity
to
kind
of
get
in
and
hear
about
stuff
that
that
red
hat
and
other
companies
are
doing
and
what's
happening
so
moving
on
I'm
going
to
introduce
our
Our
Guest
Natan,
he
is
CEO
of
a
company
called
Robusta
robusta.dev
and
before
I.
Have
him
talk
more
about
what
Robusta
is
and
does,
which
is
things
that
I'm
very
excited
about,
because
it's
in
get
Ops
and
observability
and
it's
open
source?
B
B
A
So,
thank
you
very
much
for
having
me
and
it's
especially
exciting,
like
you
said,
to
be
here
on
episode.
Number
42.
I've
also
always
been
a
fan,
though
I
didn't
bring
my
own
towel
nice
reference
there
in
the
chat
and
for
us
the
question.
I
guess
is:
what
do
you
do
when
stuff
goes
wrong
in
your
grouping
on
these
cluster
and
how
do
you
debug
stuff
in
2022?
So
if
we
look
at
them,
many
of
us
are
already
in
kubernetes
clusters.
A
A
Can
we
take
the
knowledge
and
take
it
out
of
my
head
where
I
know
how
to
investigate
a
certain
other
or
I
know
what
that
other
means
or
I
go
and
I
read
a
Blog
online
about
how
to
investigate
it,
and
can
we
put
that
knowledge
really
into
github's
file
in
a
repository
and
turn
that
knowledge
into
account?
So,
oh
that
Robusta
was
born.
B
I
mean
that
is
a
story.
An
Origin
story
that
I
I
definitely
personally
identify
with
the
the
cool
thing
to
me
right.
The
thing
that
I
I
thought
was
the
coolest
was
that
you
not
only
have
you
you
created
a
platform.
I
think
you
mentioned.
You've
got
at
least
at
least
100
users
that
you
know
about
because,
but
the
other
half
of
that
is
right.
A
Okay
so
I'll
start
with
just
to
worry
about
running
an
open
source
company
or
an
open
source
based
company,
and
the
time
has
never
been
better
for
that,
because
there's
really
a
number
of
large
companies
now
are
very
successful.
Companies
like
hashicor,
like
that's
like
rather
they've,
proven
that
ammo
can
work
and
rather
has
a
little
bit
of
an
outlet
buyer
in
terms
of
the
business
model
and
in
starting
off
more
of
a
support
based
strategy.
A
So
there
are
a
lot
of
other
companies,
though,
that
really
shown
to
VCS
and
shown
to
the
world
that
something
that
you
can
do,
and
you
can
do
it
successfully
and
even
build
a
huge
company
around
that.
So
there's
never
been
a
better
time
to
start
a
company
that
has
a
heavy
open
source
Park.
Now
speaking
about
what
we
do
know
really
the
core
of
what
we
do
is
three
things.
A
First
things
we
have
an
engine
that
listens
to
all
the
events
that
happen
in
your
kubernetes
clouds
and
you
define
in
the
ammo
triggers
and
actions.
So
you
can
Define,
for
example,
when
this
trigger
happens,
then
this
is
the
action
I'm
going
to
take
to
investigate
back
or
to
fix
it
and
I'll
show
you
some
really
specific
examples
in
just
a
moment.
The
second
part
of
it
is
I
mean
an
engine
is
no
good.
A
If
you
don't
have
any
rules,
so
we
don't
want
to
just
give
you
an
engine
and
then
say:
okay
now
go
put
together.
Your
own
recipes
put
together
your
own
rules.
Do
all
of
that.
What
we
want
to
give
you
is,
we
want
to
give
you
the
whole
experience,
so
we
want
to
give
you
an
out-of-the-box
solution.
That
knows
how
to
take
the
common
issues
in
your
kubernetes
cluster,
and
then
it
knows
how
to
tell
you
for
each
of
those
issues
what
to
do
and
how
to
fix
it.
A
So
that's
the
second
part,
and
then
the
third
part,
which
is
the
only
part
that's
not
open
source,
is
we
have
an
optional
add-on,
stacked,
app
add-on,
sorry
UI
that
runs
in
the
web
and
that
SAS
platform
gives
you
a
single
pane
of
glass.
Where
you
can
see
all
your
clusters,
you
can
see
what's
going
on
with
them
and
you
can
troubleshoot
your
shoes
from
there
by
enough
talking.
I.
Think
I
should
just
show
you
guys
the
code,
so
let's
Dive
Right
into
an
example
and
I
think
it
will
only
make
it
sense.
Then.
B
A
So
I
want
to
start
with
a
really
simple
example:
here's
an
example
that
we've
all
seen
we
have
crashes
when
the
Pod
crashes,
I
guess
the
question
is:
what's
the
first
thing
that
you're
going
to
do?
If
you
decide
that
you
want
to
go
and
investigate
that
me,
I
guess:
I
can
I
know
me
personally.
What
I
would
always
do
is
I
would
go
and
I
would
start.
I
would
then
go
and
run.
Cube,
CTL
dogs
and
I
would
fetch
those
logs
and
look
at
the
dogs
to
see
if
there's
anything
there.
A
So
here's
another
nation
that
all
meets
that
exact
process.
So
what
we're
doing
here
is
we're
defining
the
trigger
the
trigger
here
is
a
Prometheus
of
art.
There's
no
need
to
invent
the
wheel,
I
mean
you
read
the
apparently,
so
that's
firing.
Let's
take
your
existing
Prometheus
alerts
now,
let's
hook
them
up
to
Robusta,
so
when
in
Prometheus
that
they're
fires
and
that's
the
other
for
crashing
pods,
that's
the
trigger
and
the
action
is
just
to
fetch.
The
dogs
now
what's
going
on
here
is
when
this
fires
we're
kubernetes
native.
A
So
this
is
the
trigger
and
that
trigger
has
data
about
the
Pod
so
for
Matthew.
So
they're,
that's
firing,
that's
related
to
a
pod,
and
you
can
map
that,
according
to
the
labels
on
that
Prometheus
alert
that
goes
from
Prometheus
tether
manager
to
Robusta
and
when
Robusta
gets
it,
it
has
a
powered
object
and
knows
which
path
is
crashing
and
therefore
you
can
just
Define
here
in
action
that
says:
okay,
let's
fetch
the
dogs
from
that
pot.
A
And
let
me
show
you
what
that
looks
like.
So,
if
I
jump
over
here
to
my
stack,
then
here's
just
an
example
that
just
arrived
from
My
Demo
cluster.
So
you
can
see
here,
here's
the
other
and
you
can
see
right
here,
I'm
just
pulling
in
the
logs.
We
have
here
the
words
longest
Java
exception,
so
it's
a
bit
of
a
contrived
example
and
we're
pulling
in
that
data.
So
you
can
see
here
exactly
why
this
path
is
crashing.
A
Now
this
specific
issue,
so
is
just
one
example
of
one
rule,
so
I've,
actually
our
automation
here
that
will
also
pull
in
some
extra
data.
So
you
can
see
we're
actually
pulling
in
the
pad
events
as
well.
So
that's
like
running
Cube
cattle
guide
events
and
then
filtering
out
back
to
the
events
that
are
relevant
to
that
pod.
And
then
we
attach
that
extra
data
here.
B
Yeah,
it
actually
is
the
first
time
you've.
You've
walked
me
through
a
full
example.
I
will
say,
like
you
know
what
you
said:
we
basically
automated
and
put
into
chat
Ops
a
very
normal
workflow
right
of
I.
The
workflow
you've
described
I
I
have
done
exactly
by
hand,
I
I
think
the
only
difference
is
every
time
I
do
it.
It's
OC
instead
of
cube,
cuddle
or
cube
CTL,
depending
on
your
flavor
for
for
saying
that,
and
so
I'm
gonna
lie.
I
had
a
moment
where
my
brain
froze.
B
I
have
had
stories
of
being
tired
and
and
mucking
about
on
my
own,
like
kind
clusters
or
whatever,
and
wondering
why
my
commands
were
not
working
only
to
realize,
like
15
minutes
later,
that
that
I
was
using
OC
instead
of
instead
of
the
the
cube
API
directly
and
that
that
was
why
my
stuff
didn't
work.
So
for
those
of
us
who
who
have
open
shift
on
the
brain
yeah
you,
you
know
mental
mental
replacement.
A
B
Everything
everything
is
fine,
but
yeah.
This
is
a
workflow
I.
My
team
I
mean
I
got
we
do
it
almost
every
single
day
right
so
to
see
it
to
see
it
just
put
into
a
slack
for
us
that
can
save,
and
we
were
talking
about
needing
to
resolve
an
incident
right
in
the
meantime
to
resolve
that
can
save
as
much
as
five
to
ten
minutes
in
in
actual
manual
work,
and
so
that
I
think
you
didn't
mention
it.
B
A
Yep
so
then
moving
on
like
taking
them
and
expanding
on
it,
then
the
obvious
next
question
is
okay.
Well,
can
we
also
do
Channel
apps
and
can
we
fix
stuff
right
and
here
they're
actually
an
inherent
conflict
with
get
Ops,
because
when
you
fix
stuff,
then
you
don't
want
to
fix
it
either
in
your
cluster,
really
the
way
that
you
want
to
fix
it
is
you
want
to
go
and
open
like
a
PR
to
your
GitHub
repo
and
then
change
it
there.
A
So,
if
you're
going
to
go
and
fix
stuff
in
cluster,
then
you
really
here
we're
now
we're
starting
to
break
a
little
bit
to
get
up
smile.
So
we
see
here
two
philosophies,
philosophy
number
one
is
everything
should
go
through,
get
UPS,
no
matter
what
and
really
like
philosophically
I
think
that's
the
correct
way
and
that's
the
direction
that
we're
going
to
be
moving
in
with
these
type
of
remediations
that
we
implement.
A
We
don't
have
that
yet
today,
but
in
the
interim,
what
we
do
have
is
we
do
that
you
make
these
temporary
temporary
fixes
in
your
cluster
directly
from
slack.
So
you
can
click
a
button.
In
slack
and
automatically
that
will
go
and
it'll
make
a
fix
on
your
cluster
and
that's
a
very
good
for
workarounds.
Let's
say
you
want
to
restart
a
pod
or
let's
say
you
wanna
drain
the
node
or
do
something
very
specific
alert,
but
isn't
really
a
git
apps
type
of
change
anyway.
A
Here's
just
one
example
like
that
that
you
see
so
here
the
HP
reach
the
maximum
number
because
see
all
right,
quick
workaround
that
you
can
do
you
just
push
that
button
replicas
by
30
percent
and
then
you
are
going
to
want
to
go
and
commit
that
to
get
or
in
the
morning
you're
going
to
want
to
go,
investigate
that
and
do
that
properly.
But
you
get
you
get
that
immediate,
quick
mandate
that
sometimes
lets
you
maintain
those
slas.
B
Yeah
and
that's
that's
really,
that's
really
the
balance.
We
have
to
strike
right
and
you
know
Red
Hat.
We
we
have
get
Ops
for
certain
types
of
incidents,
specifically
around
anything,
about
capacity
to
do
exactly
this
type
of
thing,
but
you
know
we
had
to
create
the
the
get
Ops
tooling
and
Integrations
to
do
that,
because
before
that
we
were
having
to
do
it
by
hand
and
then
retroactively
like
go
back
and-
and
you
know
it's
the
equivalent
of
remembering
to
go
back
and
update
the
documentation
after
you've
made
the
change
right.
B
A
B
A
B
I
think
that's
a
really
good
way
to
put
it
so
yeah.
It's
that's
and
I
I
really
agree
with
you
on
when
you
went-
and
you
said,
like
there's
kind
of
like
that
that
dichotomy
of
you
have
to
fix
the
issue
so
like
chat,
Ops
allows
you
just
like
you
know,
quickly:
slap
the
Band-Aid
on
and
the
other
part
is.
You
have
to
have
record
of
what
you
did
to
fix
the
issue,
and
so
that's
why
git
needs
to
be
the
sort
of
Truth
and
and
yeah.
B
It's
I
I
think
whenever
we
can
use
get
get
Ops
to
do
incident
response,
we're
better
for
it
and
if
there's
a
problem
in
our
incident
response,
if
you
get
Ops,
then
we
can.
You
know
life
cycle
and
maintain
that
we
can
break
that
into
the
testing
cycle
of
the
product.
Even
so,
you
can
always
basically
guarantee
your
operations
by
doing
operations
is
code,
which
is
actually
one
of
the
big
cultural
changes
that
we've
been
working
on.
B
Implementing
in
red
hat
is
getting
getting
all
of
our
Ops
pieces
actually
into
the
QE
cycles,
and
we've
had
some
pretty
excellent
Partnerships
from
our
QE
teams.
To
do
this
because
Red
Hat
has
an
initiative
called
operate
first,
our
QE
teams
have
been
pretty
heavily
involved
in
that,
in
addition
to
some
of
our
developers
and
some
of
our
operational
groups
to
kind
of
create
this,
this
concept
that
you
need
to
know
your
day
two
operations
and
a
lot
of
times
day.
A
Yeah
yeah,
so
that's
that's
a
big
part
of
it.
I
mean
big
part
of
it
is
really
taking
that
knowledge
that
you
have
like
I've.
Seen
that
happen,
where
I'm
eating
dinner
with
someone
and
in
the
middle
of
eating
dinner,
like
a
friend
of
mine,
gets
called
on
the
phone,
and
he
has
to
explain
what
exactly
what
commands
you
should
run
and
say:
okay,
go
run
this
command
and,
like
it
take
like
almost
reading
out,
not
everybody
better.
B
I
also
love
that
it's
done
with
yaml
right,
because
at
this
point,
yaml
is
becoming
like,
like
almost
like
a
it's
just
like
a
language.
We
all
know
not
even
language,
but
you
know
what
I
mean
right.
It's
like
it's
like
it's
like
a
language.
We
speak
right.
We
can.
A
lot
of
us
can
lint
yaml
in
our
sleep
basically,
so,
for
you
know
that
kind
of
normalization.
That's
the
word.
B
A
B
A
Answer
the
interesting
one,
because
right
ansible
is
configured
in
yaml,
but
then
you
extend
it
in
Python,
right,
yep,
now's,
a
good
time
for
me
to
speak
about
like
what
exactly
are
these
actions
right,
because
it's
they
actually
the
exact
same
model?
Okay.
So
if
I
want
to
break
this
down
again
and
like
every
automation
that
you're
going
to
Define
in
Robusta
has
a
trigger,
has
an
action,
and
it
has
the
sync.
The
trigger
is:
what's
triggering
this
automation
to
kick
off
in
your
kubernetes
cluster.
A
The
action
is
what
data
you're
Gathering,
or
what
automatic
or
semi-automatic
remediations
you're
suggested
to
the
user.
So
what
buttons
are
you
giving
them
in
stock
and
whenever
I
say
stack
like
if
you
work
at
a
company
with
over
a
thousand
people
or
over
10
000
people,
just
replace
them
you're
added
with
Ms
teams,
so
it's
stack
or
whatever
other
messaging
app
that
you're
using
so
and
then
the
sinks
is
the
destination
that
you're
sending
that
to.
A
So
if
we
look
at
this
now,
I'm
going
to
break
that
down
a
little
more
and
show
you
what
each
of
these
parts
really
is.
So
if
I
jump
over
to
our
documentation,
then
I
can
just
go
down
here,
and
here
we
have
a
section
on
triggers
and
if
I
click
around
the
triggers
now
you
can
see
all
the
different
triggers
that
you
can
bring
to
Robusta
so
from
Prometheus
standard
manager.
That's
what
we
were
looking
at
so
far
choose
from
the
API
server.
A
So,
for
example,
if
a
deployment
is
updated,
maybe
you
understand
and
know
this
in
the
slack
Channel
10
minutes
after
the
point
was
updated
and
say:
okay,
like
here,
are
the
graphs
how
the
application
is
performing
after
the
latest
update
and
then
various
different
triggers
there.
So
cluster
and
under
the
hood,
that's
powered
by
what
was
one's
Cube
watch,
bitnami
developed,
Cube
watch,
then
Vietnam
got
acquired
by
VMware.
Then
the
project
was
abandoned.
Now,
if
you
go
Q
watch
on
GitHub,
it
refers
to
us.
A
We
took
over
the
official
fork
for
that.
So
that's
what's
powering
this
under
God
and
you
can
so
you
have
various
triggers
you
have
here.
The
triggers
like
Prometheus,
like
kubernetes,
can
do
Triggers
on
the
schedule.
You
can
do
triggers
that
you
trigger
manually
by
running
command
line
or
by
webhook.
You
can
take
triggers
from
elasticsearch.
So
all
these
different
Integrations-
and
that
just
is
the
first
part
over
here-
the
trigger
for.
A
What's
going
to
kick
off
this
Automation
and
again,
I
want
to
emphasize
what
it
means
here
to
be
kubernetes
native
is
that
each
one
of
these
triggers
has
a
kubernetes
object,
that's
attached
to
it.
So
if
a
trigger
was
a
deployment
that
updated,
then
there's
a
deployment
object
attached
to
that
that's
flowing
through
this
Pipeline
and
the
action
will
get
is
input
a
kubernetes
deployment.
If
here
you
have
a
Prometheus
alert,
that's
firing
on
a
pod
and
it's
mapped
to
that
part
by
the
alerts.
A
A
So
now
I'm
going
to
jump
over
to
these
actions,
so
you
have
a
trigger
fired
off
a
trigger
with
a
kubernetes
object,
that's
associated
with
it
and
then,
if
we
go
over
here,
then
there
are
all
these
different
built-in
actions.
So
there
are
actions
that
will
grab
a
graph
from
grafana
and
render
it
there
are
actions
that
will
take
an
action
or
allow
you
to
take
action.
When
you
click
a
button,
there
are
actions
that
we're
pulling
the
green
on
these
events.
A
There
are
actions
that
will
run
the
CPU
profiler
on
the
Java
application,
their
actions
that
will
grab
a
stack
Trace
from
a
Python
program.
That's
running
inside
that
part.
So
there
are
all
these
different
debugging
actions
that
can
gather
this
wide
range
of
data.
I
want
to
show
you
what
those
actions
really
are.
So,
if
I
take
again
this
logs
in
retraction
I'm,
just
going
to
jump
over
to
the
source
code
for
it,
it's
just
a
python
function
and
I
want
to
show
that.
A
Okay,
so
you
see
here
this
action,
that's
the
action
that
we
configured
in
the
ammo
and
it
takes
two
inputs.
So
the
first
input
here
is
the
power
of
x
and
not
eventually,
that's
me
get
the
Pod
itself
and
fetch
in
the
other
metadata
now
I
have
here
the
pound
object
and
I
can
just
do
anything.
I
want
now
in
this
kubernetes
pad,
so
I
can
access
all
the
ammo
fields.
A
A
A
So
it's
a
very
easy
reaction
to
this.
That
men
have
logic,
that's
even
specific
to
your
application.
You'll,
not
check
that
specific
to
a
specific
type
of
error
that
occurs.
So,
let's
say,
for
example,
you
have
an
error:
that's
firing
on
a
bunch
of
500
errors
in
your
application.
That's
serving
HTTP
traffic
and
you
say:
okay
now,
I
want
to
go
and
I
want
to
grab
like
from
istio
the
network
tab.
A
We
data
on,
what's
going
on
and
I
want
to
attach
that
now
to
this
other
that
I
arrive
in
stack,
so
we
haven't
written
an
action
for
that
yet
I
mean
we
probably
should
at
some
point.
But
if
you
wanted
to
go
in
right
in
action
like
that
yourself,
it
really
would
just
be
a
few
lines
of
python
and
then
it's
very
easy.
We
have
an
API
wraps
the
stack
API
in
them
as
team,
so
you
just
attach
it
to
the
outgoing
message
now
I
see.
We
also
have
a
question
from
the
chat
right.
B
A
Do
we
want
to
bring
that
up
on
screen,
I
think
the
question
is
that
you
could
take
a
Vibe
of
system
and
knowledge
responses
and
consider
your
machine
learning
applications,
and
the
answer
to
that
is
absolutely
yes
and
we're
approaching
this
here
kind
of
in
the
opposite
direction.
I
think
there
are
some
companies
that
say:
okay,
we're
gonna,
gather
all
this
data
and
then
we're
going
to
throw
a
machine
learning
algorithm
at
it.
Yep
it's
a
very
cool
approach.
We're
approaching
that
from
actually
the
opposite
direction.
A
We're
saying:
okay,
we're
going
to
gather
all
this
data,
we're
going
to
hand,
write
the
first
100
200
responses
and
we're
going
to
really
for
each
one
we're
taking
another
we
go
on.
We
go
on
stack
over
without
we
research
it.
We
read
through
the
kubernetes
documentation,
we
look
through
the
kubernetes
source
code
and
we
try
and
figure
out
for
each
common
error
and
for
each
other
that
people
tell
us
about
that's
going
on
in
their
cluster.
Try
and
figure
out.
A
What's
the
correct,
robust
way
to
solve
this,
and
what's
the
best
practice
for
this,
what
would
the
world's
best
systems
men
do?
Who's
familiar
with
njinx
when
this
engine
server
occurs
and
then
we
take
that
knowledge
and
then
we
turn
that
into
another
major
Playbook,
and
we
push
that
out
on
the
next
version
to
all
the
users
of
robust
assuming
they've
opted
in
to
get
those
updates.
So
we're
not
doing
it
today
with
machine
learning,
I
think
the
real
valuable
part
of
this
is
you're.
A
Building
off
you're
building
out
this
very
valuable
data
set
and
you're
building
up
the
sample
size
and
you're,
also
building
out
kind
of
this
training
set
for
what
responses
then,
like
people
use
or
don't
use,
assuming
that
they
opt
into
sending
data
to.
We
actually
don't
gather
that
data
Today
We
Gather
very,
very
little,
and
you
can
disable
to
that
material
together.
A
Then
you
have
all
this
architecture
and
things
and
models,
and
you
can
put
it
on
top
of
that
and
yeah
I-
think
the
advantage
of
starting
off
with
more
of
a
signature
based
approach
is
that
you
really
get
to
do
stuff.
That's
really
really
fine-tuned.
Like
I
packed
our
discussion
on
CPU
limits.
What
kicked
this
all
off?
Was
we
actually
wrote
a
robust,
the
remediation
that
said?
Okay,
you
have
CPU,
throw
a
thing:
what
should
the
proper
fix
for
that
be?
A
B
These
types
of
things
that
you
have
to
keep
in
mind,
depending
on
literally
all
the
way
down
to
like
your
Hardware
situation,
so
I
I
think
I
I
do
I,
do
want
to
I
I,
do
agree,
there's
lots
of
opportunities
for
for
limits
to
not
become
necessary
and
then
there's
times
where
I
think
the
SCP
limits
can
be
very
important.
And
of
course
you
don't
want
to
overclock
your
physical
CPU,
if
you
don't
have
to
now.
If
you
want
to
overclock
somebody
else's
CPU
because
they're
paying
for
it.
The
hacker
in
me
supports
this.
A
Okay,
so
that's
a
really
cool
example,
because
you
know
what
I
love
about
that,
though
it
shows
something
that
we
deal
with
all
the
time,
which
is
that
the
answer
to
specific
alert
is
it
depends
very
very
often
so.
For
example,
I'll
give
you
like
a
really
specific
instance.
When
there's
CPU
thriving
you
want
to
ignore
it.
A
If
you're
running
on
gcp
and
the
CPU
throttling
occurs
on
Metric
server,
then
you
want
to
ignore
that
CPU
throwing
for
throttling,
because
there's
no
way
in
the
world
that
you
can
fix
it
like
metric
server
is
managed
by
gcp
by
the
Google
kubernetes
engine.
And
if
you
change
like
the
CPU
limits,
you
change
anything.
A
It
just
rolls
back
the
change,
but
I
can
tell
you
that
anyone
who
installs
Prometheus
via
chart
on
on
the
kubernetes
by
default
every
single
person
you
use
its
default
alerts,
has
that
other
firing
and
taking
up
room
in
their
stack
Channel
and
ultimately,
you
get
this
other
critique.
And
you
stop
looking
at
the
alerts
because
there's
this
stuff
in
your
stock,
Channel
That's,
Just,
Pure,
garbage
yeah
I,
can
tell
you,
though,
anyone
who
takes
the
number
one
Helm
chart
for
installing
Prometheus
on
kubernetes.
A
Anyone
who
takes
that
will
have
this
garbage
alert,
firing
all
the
time
and
there's
nothing
you
can
do
in
the
world
to
fix
it.
So
now,
if
you
add
in
a
tool,
then
like
Robusta,
we
actually
do.
Is
we
actually
take
that
specific
alert
and
we
have
code
in
there?
That
checks.
Are
you
running
on
gcp?
Is
this
alert
firing
a
metric
server
and
if
it
is,
then
we
just
silence
the
other,
and
we
don't
bother
sending
it
to
you,
because
there's
nothing
you
can
do
about
that
in
the
entire
world.
A
B
B
This
is
something
that
we
battle
every
single
day,
that
that
exact,
like
almost
mental
exhaustion,
from
being
overwhelmed
by
alerts
that
don't
matter
you
can't
fix
I,
do
want
to
go
back
to
the
machine
learning
concept
a
little
bit,
though,
because
this
is
interesting
and
I've
actually
recently
looked
at
some
tools
that
do
apply.
Machine
learning
and
I
I
really
feel
like
there's
there's
a
crawl,
walk,
run
approach
here
and
I
think
that
if
you
want
to
apply
machine
learning,
you
need
to
deeply
understand
your
environment
and
what
your
data
truly
means
before.
B
You
can
actually
start
really
training
a
machine
learning
model.
So
what
you
were
talking
about
earlier
in
a
ton
about
how,
like
all
that
manual,
fine
tuning
I,
think
that's
that's
really
important.
When
we
start
looking
at
some,
almost
every
workload,
Red
Hat
sres
deal
with
is
a
kubernetes
operator
based
workload,
and
that
is
kind
of
the
same
thing.
We've
taken
that
operational
system
and
knowledge,
we've
codified
it
into
a
kubernetes
operator
which
will
it
reconcile.
So
the
operator
has
operands.
B
Interesting
keep
going.
Oh
yeah,
I
love,
I
love
to
talk
about
that.
So
an
operator,
it
extends
the
kubernetes
neti's
API
extends
the
the
ability
to
have
a
stateful
environment,
and
it
actually
basically
puts
that
on
the
application
layer
through
the
use
of
custom
crds,
and
so
the
kubernetes
operator
watches
its
operands.
Those
are
the
kubernetes
objects
that
it
needs
to
keep
in
touch
and
it's
expecting
a
certain
state
for
those
operands
right.
B
This
is
this
is
codified
human
knowledge,
but
they
are
very
robust
pieces
of
software
themselves
and
they
do
fail
as
well.
In
fact,
the
talk
that
I'm
going
to
be
giving
at
Dev
comp
is
what
I've,
given
before
a
couple
times
called
Helm
and
back
again
and
I,
actually
talk
about
some
of
the
failures
and
problems
with
the
kubernetes
operator
model
like
the
some
of
the
gotchas,
when
I
do
a
comparison
between
those
two,
those
two
tools
and.
B
I
can
send
you
the
link
to
the
previous,
the
I
I.
Did
it
for
Dev
confero
in
January
and
I
need
to
update
it
a
little
bit
because
there
have
been
new
features
added
to
helm.
That
I
want
to
account
for
I
want
to
get
deeper
into
the
ansible
operator,
because
you
can
do
a
goaling
operator
or
an
ansible
operator,
and
since
the
operator
SDK
is
ansible
under
the
hood
I
like
just
using
the
Straight
ansible
like.
B
B
The
operator
codifies
the
human
knowledge,
so
it
kind
of
it
kind
of
I
want
to
say
usurps
the
need
for
machine
learning
on
a
lot
of
these
to
a
degree
and
then
into
the
operators.
Right.
Operators
have
a
maturity
model
between
one
and
five
and
a
level.
Five
operator
is
amazing,
because
that
operator
will
Auto
scale
your
application,
there's
deep
insights,
additional
monitoring
and
alerting
and
metrics,
and
everything-
and
all
of
this
is
of
course
kept
in
git.
A
A
Is
they're
running
this
reconciliation,
Loop
right
and
you're,
comparing
your
SPAC
so
you're,
comparing
like
status
where
you
are
right
now
to
the
specking,
where
you
want
to
be
and
then
you're
bringing
those
to
in
mind
and
I
always
think
about
this.
In
the
way
the
kubernetes
documentation
describes
it,
which
is
in
terms
of
Robotics
and
control
loops.
When
I
was
in
high
school
I
participated
in
robotics
competition
and
like
the
hardest
thing
we
did
was
to
try
and
get
the
robot
to
drive.
A
So
what
you
always
do
is
every
control
algorithm,
typically
PID,
when
you're
in
high
school
and
you're
looking
at
like.
Where
are
you
proportionate
to
where
you
want
to
be
or
looking
at
the
integral
or
looking
at
the
derivative
and
trying
to
bring
yourself
back
in
line
and
like
term
after
right
and
adjust
the
speed
on
each
motor?
So
that's
how
I
view
like
kubernetes
operators,
I
mean
I,
don't
know
if
you
view
it
the
same
way,
but
that's
at
least
like
my
mental
model,
that
I
sort
of
have.
B
That
is
a
great
mental
model.
I
think
yours
is
better
and
more
accessible
than
mine
is
I've,
always
just
been
like
man.
It's
a
lot
like
a
like
an
old
school
demon
right,
just
like
a
Linux
demon.
It's
just
continuously
running
software.
That
does
some
things,
which
is
like
the
most
like
basic
and
reduced
down.
Explanation
for
what
an
operator
is
so
I
think
yours
is
better
and
I'm
going
to
steal
it.
Okay,.
A
So,
what's
cool
about
operas
is
that
it's
not
event
driven
meaning
like
you
to
some
degree,
you
don't
care
about
the
history
and
again,
maybe
this
is
different
for
certain
types
of
operators
like
if
you
look
at
even
the
built-in
controllers
or
like
a
demon,
sat
right
like
or
a
deployment
it
doesn't
care.
If,
five
minutes
ago
you
had
like
10
replicas,
and
now
you
have
five
like
you're,
just
calculating
at
any
given
moment
in
time.
A
That
makes
it
really
easy
also
to
write
robot
software
because
you're
always
recalculating
you're,
always
adjusting
you
don't
need
to
care
about
all
that
history,
but
from
an
observability
perspective.
Sometimes
that's
really
like
yeah.
Let's
say
you
have
a
pod
deployment
and
it's
crashing
constantly
and
every
time
it
crashes
then,
like
your
customers
have
a
horrible
experience
and
you'll
deeply
I
keep
going
on
LinkedIn,
getting
like
a
blank
page
and
have
to
restart
it.
I
get
these
500
errors
like
every
time,
you're
going
You're
browsing
somewhere.
A
A
It's
really
really
awful,
like
in
kubernetes,
you
like
to
say,
like
you're,
think
of
pods
and
your
workloads
as
cattle
like
not
pets,
but
if
all
your
like,
if
your
entire
flock
is
dying
or
like
50
of
them
are
dying
and
then
coming
back
to
life
and
they
never
get
it
past
H5,
like
you,
have
a
problem
there
so
yeah
the
operator
perspective
is
really
really
useful
on
accident.
If
you're
building
great
robust
software,
but
you
when
you're
trying
to
do
observability,
you
actually
want
something.
That's
event
driven
so
yeah.
A
That's
the
difference
like
I,
wouldn't
try
and
build
any
sort
of
operator
with
Robusta,
because
when
you're
writing
something
that's
like
comparing
us
back
in
state,
you
don't
want
something,
that's
event
driven,
but
when
you're
doing
observability
actually
what's
event
driven
in
the
state
changes
they
really
really
matter.
Yes,.
B
B
I
love
talking
with
a
person
because
it
ends
up
conversational
like
this,
and
we
can
like
teach
each
other
things.
We
really
maybe
didn't
even
know
we
knew
did
or
didn't
know,
rather
so
I
I,
like
I
I,
don't
actually
like
talking
alone.
I
hate
it,
but
I
actually
have
a
couple
production
stories
to
exactly
illustrate
the
points
that
you
said.
B
Course,
okay,
so
the
first
one
right
that
it's
it's
not
event
driven
and
that
is
I
I
frequently
call
the
reconcile
Loop
in
an
operator
a
double-edged
sword,
because
sometimes
you
just
want
things
to
fail
because
you
need
to
know
they're
failing
and
the
operator
will
hide
failures
by
reconciling
States.
So
you
can
have
something
crashing
and
you
know
you
might
not
necessarily
always
catch
it.
You
can.
B
You
can
actually
tune
that
in
the
observability
so
that
you
can
catch
if
something's
crashing
and
restarting
a
lot
and
there's
there's
a
lot
of
ways
to
to
to
to
tune
that
in
and
be
in
in
SRE.
We
actually
spent
a
lot
of
time
working
on
making
sure
alerts
and
warnings
are
tuned
and
resolved.
So
that's
cool.
B
Yeah,
it's
just
actually
another
implementation
of
Prometheus
and
alert
manager,
yeah
and
there's
a
few
other,
a
few
other
toolings
in
there
so
yeah.
So
that's
a
double-edged
sword
that
reconciled
Loop
right,
because
sometimes
you
want
a
failure.
You
want
to
know
that
it's
failing
right
and
it's
it's
not
it's
not
necessarily
revealing
that.
There's
a
problem
right,
a
pod
might
get
up
start
running,
run
for
five
minutes
and
crash
and
it
restarts
well
if
it
restarts
fast
enough.
B
Some
of
your
alerting
is
not
going
to
catch
that
this
is
that,
there's
that
there's
an
issue
right
and
you
end
up
finding
out
from
the
customer
that
that
definitely
has
happened
a
lot.
The
other
thing
is,
is
you
talked
about?
It
doesn't
know
past
numbers
like
that
five
minutes
ago
there
were
15
pods
instead
of
the
five
it's
expecting.
This
actually
just
happened
to
us.
A
B
Basically,
there's
there's
this
there's
this
pod.
That
runs
a
controller.
The
controller
says:
okay,
based
on
XYZ
factors,
I
need
Z
number
of
PODS
to
do
some
networking
stuff
right,
and
so
it
creates
the
number.
So
there
was
a
problem
in
that
math
and
instead
of
creating
two,
it
was
creating
22
and
this
actually
came
from
the
open
source
Upstream.
B
Well
then,
the
operator
caught
that
that
was
too
many
and
reconciled
it
back
down
to
the
correct
number
and
because
the
operational
knowledge
was
baked
into
the
operator,
whereas
the
controller
that
you
know
was
installed
by
the
software
had
had
a
mathematical
error
and
then,
when
that
happened,
the
other
posit
had
been
killed.
All
came
back
up,
so
we
were
seeing
up
and
down
outages
in
production
right
and
we're
like.
Why
is
this?
And,
of
course,
you
know,
there's
a
Cron
job
at
fault.
B
Here,
there's
a
crown
job,
that's
restarting
some
stuff
or,
like
you
know,
checking
some
stuff
and
starting
starting
up
starting
up
the
process
to
calculate
every
so
often.
So
it's
always
a
crown
job,
I
swear
and
so
that
that
it
was,
it
was
terrible
because
the
operator
kept
fixing
it.
It
took
three
sres
all
watching
different
parts
of
the
cluster
to
see
wait,
a
minute
that
went
up
to
22
and
then
went
back
down
to
two
in
within
the
five
minute
reconcile
Loop
and
that's.
Why
we're
having
all
these
problems
and.
B
A
One
of
these
days,
what
I
really
want
to
do
is
I
wanted
to
do
a
kubernetes
like
flight
recorder,
where
you
take
a
given
object
and
then
just
track
all
the
various
changes
that
undergoes
and
why
and
it's
not
hard
to
do
like
I,
could
it
it
would
be
fairly
trivial
to
even
right,
but
just
to
give
you
like
that
whole
story.
We
do
some
stuff
in
Robusta
that
touches
on
it,
but
it's
really
aimed
at
a
different
use
case,
not
for
the
use
case
where
you're
writing
operators.
B
A
Cool
yeah
I
have
to
check
that
out
mate.
So,
but
you
touched
on
live
stuff
here
that,
like
I'm,
really
passionate
about
that,
we
end
up
speaking
about
that,
because
I
think
that
case,
where
you
have
a
part
that's
going
up
and
down,
we
see
that
all
the
time
with
them
kills.
Someone
says:
okay,
I
have
a
new
kill
and
you
look
at
the
graph
then,
and
you
see
them
killed.
Like
looks
like
this,
the
graph
of
the
memory
usage,
maybe
I,
can
even
show.
Let
me
see
if
I
can
show
a
specific
example.
B
B
So
well,
I'll
take
advantage
of
this
a
little
bit
low,
because
I
think
we're
gonna
go
right
up
to
the
end
today,
because
of
how
just
amazing
this
content
and
conversation
is
to
remind
everybody
to
to
like
And
subscribe
to
the
Channel,
please
so
that
they
will.
Let
me
keep
doing
this
that's
very
important
and
then,
if
you
don't
already
follow
me
on
Twitter,
follow
me
on
Twitter,
I'll,
post,
the
helm
and
back
again
talk
the
Youtube
video
for
it
and
I
will
go
ahead
and
Natan.
B
Excuse
me:
I
keep
doing
that.
I
have
shame
on
me.
I'll
go
ahead
and
tag
you
in
it
too.
So
if
anybody
who
follows
you
might
also
see
it,
but
also
follow,
follow
him
because
he
posts
good
content
very
regularly.
You
recently
just
posted
like
a
great
like
what
are
the
best
kubernetes
for
dummies
videos,
I
loved
that
thread.
That
was
an
amazing
thread.
A
A
A
And
paper
I
I
got
what
I
was
looking
for.
Okay,
so
I
want
to
make
I'm
gonna
take
an
example
here
from
another
and
I'll
get
to
show
off
another
robust
stuffing,
while
we're
at
it
so
cool.
If
you
think
of
like
a
normal
alert
that
you
get
for
an
even
kill,
then
I
mean
I'm
sure
you
dealt
with
debugging
girls
before
right.
So
again,
something
is
killing
good
luck
and
it
doesn't
usually
have
enough
context
to
really
know
what
you
need
to
go
and
do
so
something
just
added
on
recently
to
robust.
A
That
also
demonstrates,
like
this
type
of
behavior
you
see
with
the
boom
kills,
is
when
there's
no,
then
we
fetch
the
data
from
chromatic
BS
about
this
card,
that's
been
killing,
and
then
we
just
show
you
here
like
the
memory
usage
for
that
pad
over
time,
and
then
you
can
see
right
away.
Okay,
here's
the
memory
usage.
We
crossed
this
once
and
went
up.
A
We
actually
went
way
above
the
limit
and
then
we
got
and
then
got
killed
and
we
went
down
and
we've
stayed
below
it
since
then,
so
that
has
a
different
sort
of
profile
than
something
that's
constantly
growing.
That
goes
right
back
up
afterwards
as
soon
as
it
gets
killed.
So
by
just
looking
at
graph,
you
can
sort
of
understand:
okay,
what's
the
behavior
here
and
then
what's
the
fix,
so
I
just
need
to
raise
the
limit
because
it
was
a
one-time
Spike,
whereas
there's
something
here,
that's
leaking
all
the
time.
B
Yeah-
and
it's
actually
so,
this
is
really
interesting
because
you
briefly
touched
on
on
the
characteristics
of
a
memory
leak,
right
and
I.
Think
this
is.
This
is
really
important,
so
in
in
a
lot
of
CPU
and
memory
are
often
very
like
if
one's
increasing
the
other
one
might
also
be
increasing
as
well,
but
they
can
they're
they're
completely
in
independent
and
a
great
example
of
this
is,
if
you're
doing,
video
processing
you're
going
to
use
a
lot
of
CPU
and
not
a
lot
of
memory.
B
There
are
other
things
where
you're
going
to
be
using
them
a
little
bit
more
jointly.
So
if
you're,
if
you're
getting
an
kill
on
a
memory
on
a
memory
right,
your
rest
of
your
pod,
Health
metrics,
might
be
like
totally
fine,
except
for
towards
the
very
end.
As
the
memory
starts
to
run
out,
the
CPU
will
work
harder
and
harder
and
harder
to
try
to
reclaim
any
available
memory
to
prevent
an
right.
A
B
B
A
Okay,
so
the
secret
that
I'll
say
first
of
all,
is
that
underneath
everything
I'm
going
to
show
it's
really
just
the
open
source,
so
we're
just
taking
all
the
observability
data,
that's
gathered
by
an
open
source
and
then
we're
making
an
accessible
in
one
place
and
from
all
your
clusters,
because
sometimes
when
you
try
and
consume
there's
a
stock
Channel.
Then
you
end
up
with
something
like
this.
Like
430,
plus
new
messages.
A
This
is
actually
a
channel
that
someone
else
is
running.
So
in
this
case,
it's
it's
going
back
weeks,
but
and
it's
hard
to
then
go
through
and
slice,
and
they
stand
to
look
at
that
and
you
kind
of
want
to
see
stuff
in
a
different
way
than
what
you
can
just
see
in
the
tool
like
stack
or
a
messaging
app.
A
So
that's
where
the
rest
of
the
UI
comes
in
and,
like
I
said
earlier,
the
disclaimer
here
is
the
robust,
the
UI
part
of
the
open
source.
It
does
have
a
super
extensive
retail,
so
up
to
50
nodes
free
and
for
people
who
need
they
can
also
self-host
it.
But
this
isn't
part
of
Open
Source,
just
as
a
disclaimer
I
mean
okay.
So
if
I
jump
over
here,
then
I
can
see
here.
A
A
So
let's
say
I
just
go
over
here
and
I
zoom
in
and
I'll
tell
a
good
story
here,
even
though
it's
really
not
true,
because
this
is
a
demo
environment
that
has
other
weird
stuff
going
on.
But
I'll
tell
the
story.
That's
usually
true,
so
I
come
here
and
I
see.
Okay
and
I
have
some
kubernet
these
morning.
Events
from
around
that
same
time
we're
also
pulling
in
all
the
events
of
what
you
would
get
with
Cube
Calo
gather
events
and
the
usual
data
that
you
would
gather
here
and
I
see
here.
A
This
icon,
that's
different
from
all
the
other
icons,
and
this
icon
is
a
change
that
happened
in
my
cluster,
so
we're
auditing
and
tracking
all
the
different
changes
here
that
happen
in
your
kubernetes
cluster
and
again
we're
the
official
maintainers
now
with
Cube
watch.
So
it
makes
a
lot
of
sense
and
sorry
I,
just
clicked
on
the
wrong
thing
to
reset
that
view.
So
if
I
zoom
in
now
and
I,
just
click
on
this
change
and
I
can
see
here
the
exact
change
that
happens
right
before
that
pod
started
crashing.
A
So
it's
kind
of
like
get
the
name,
but
for
your
kubernetes
cluster
and
again
this
is
a
demo
environment.
So
it
changes
again
to
actually
cause
it
to
crash
and
there's
all
sorts
of
other
weird
stuff
happening.
But
usually
you
would
see
here
like
the
new
version.
That
was
rolled
out
and
you
know
what
change
caused
this
to
happen
and
then
you'd
be
able
to
go
and
like
roll
back
that
change
or
roll
outfix.
A
So
that's
one
part
of
what
we
do
in
this
in
this
platform
and
we
pull
in
all
that
data
you
put
on
a
timeline.
You
can
see
a
timeline
for
each
for
each
specific
microservice.
You
can
see
a
timeline
for
all
the
stuff
in
your
cluster,
that
I
can
slice
and,
like
that
different
ways,
so
I
can
like
look
at
this
first
service.
I
can
look
at
this
per
particular
that's
happening
and
to
put
this
kind
of
in
more
technical
terms,
we're
kind
of
just
running
a
front
end
for
other
manager.
A
B
And
the
only
thing
I
was
gonna
ask
about
was
grouping,
and
then
you
showed
it
and
I
was
like
fantastic.
That's
like
the
thing
that
I
mean.
Obviously
you
know
that's
such
a
quality
of
life.
Piece
I
do
want
to
thank
our
hope9.
For
wishing
me
a
belated
happy
birthday,
my
birthday
was
on
Monday,
so
appreciate
the
well
wishes,
that's
very
nice
of
you
and
I.
Also
you
remembered
so
that's
also
very
nice
of
you
and
then
yes,
I
will
enjoy
my
vacation.
I
promise.
A
Yeah,
so
getting
back
here
then
the
second
you're
going
to
want
to
do
if
you
have
something
like
this
is
now
you're
going
to
want
to
drill
down
and
you're
going
to
want
to
debug
this
or
you're
going
to
want
to
look
at
this
and
understand
why
something
is
happening.
So
we
have
all
these
different
actions
built
into
robust
start
to
troubleshoot.
Right
like
we
can
gather
the
events,
we
can
run
a
profiler.
We
can
do
all
these
different
troubleshooting
actions.
A
So
what
we're
now
starting
to
do
is
we're
starting
to
expose
those
in
the
UI
as
well.
So
I
can
now
go
and
I
can
like
look
at
the
list
of
PODS.
That's
running
here
and
I
can
drill
down
and
I
can
do
various
actions
now
right
now.
The
only
two
actions
you
see
here
are
really
basic,
like
restarting
the
Pod
refetching
its
logs.
A
But
what
we're
doing
here
is
we're
going
to
be
adding
on
all
the
different
actions
in
Robusta,
which
they're
a
whole
lot
of
so
to
really
take
all
those
troubleshooting
capabilities,
and
then
it
doesn't
matter
you
can
run
them
automatically
and
gather
the
data
and
send
this
back.
You
can
run
them
when
you
need
and
going
back
to
like
a
git,
git,
apps
workflow
to
codifying
operational
knowledge
you
want
to
let
people
run
debugging
stuff,
sometimes
in
production
if
they
have
to,
but
even
when
they
let
them
run
anything.
A
B
Yeah,
that
is
super
important
in
highly
regulated
Industries.
So
people
ask
me
all
the
time
do.
Sres
at
Red,
Hat
have
cluster
admin
and
I
always
tell
them
nope
we're
running
with
a
highly
specific,
very
controlled,
our
back
and
like
exactly
like.
That
is
it's
such
an
important
thing,
but
also
that
that
type
of
thing
right,
where
you
have
some
sort
of
prove
actions
that
can
also
give
you
know
if
you're
working
at
something
like
we're
infrastructure
as
a
service
a
little
bit
more
right.
B
So
our
Cee
people
have
access
to
our
monitoring,
alerting
and
metricing
and
so
forth,
so
that
that
kind
of,
like
you
can
do
these
things
and
get
you
you
have
this
approved
action
of
get
logged
right.
That's
the
thing
that
our
Cee
folks
will
often
do,
and
so
that
the
importance
of
that
control
in
your
UI
cannot
be
overstated
because
of
of
like
the
scope
of
industries
that
are
using
kubernetes
these
days
and
are
using
kubernetes
with
SRE
support
and
and
so
forth.
A
Yeah
for
for
us,
that's
the
selling
point
A
lot
of
times
when
you
see
these
large
organizations.
So
if
it's
a
Fortune
500
company,
then
they're
going
to
want
something,
that's
really
different
than
then.
Why
a
small
startup
once
and
then,
when
you
build
out
also,
we
spoke
a
little
bit
at
the
beginning
about
the
economics
of
Open
Source.
A
But
then,
when
you
build
out
your
your
business
strategy
and
you
build
out
even
the
way
you
price
a
product
and
you
build
out
all
that,
then
like
there's
a
question:
do
you
focus
on
some
of
the
small
companies
or
you
focus
on
larger
companies
and
at
least
for
us,
we're
focused
on
the
larger
companies?
But
you
know
this:
like
small
companies
sometimes
need
some
of
this
too,
or
some
more
companies
want
to
work
the
best
way
and
even
if
you
try
and
monetize
it
you're
not
going
to
get
that
money
out
small
startups.
A
So
I
was
recently
speaking
to
the
advice
of
post
hog
and
he
said
something
that
I
really
really
liked,
which
is
like
try
and
monetize.
Don't
try
and
make
money
from
startups
so
like
if
you're
building
out
all
this
tooling,
then
the
approach
that
we
think
is:
okay,
we'll
give
it
to
people
for
free
and
you
can
kind
of
start
with
the
smaller
stuff.
That's
appropriate
for
you
and
I
see
the
commentary
also
about
having
the
right
tools
and
that's
very
true
as
well.
A
B
A
Yeah
well
wow,
it's
actually
fairly
easy.
You
just
have
to
set
up
a
printful
account
like
I
have
here
the
the
new
Robusta
t-shirt
and
you
have
on
one
of
the
older
Robusta
t-shirts,
and
this
one
says:
I
love,
open
source,
open
source
loves
me,
and
yours
is,
of
course,
upon
on
CPU
limits.
This
is
my
kubernetes
has
no
limits,
so
we
actually
print
out
like
a
whole
lot
of
these.
Sometimes
it
feels
like
we're
almost
the
number
one
kubernetes
T-shirt
shop
in
the
world.
B
No,
so
the
original
Red
Hat
was
actually
this.
The
the
guy
just
wore
a
red
baseball
cap
and
I'm
gonna
assume
so
bad
right
now,
because
his
name
just
escaped
my
brain
and
it's
like
how
how
can
I
not
come
up
with
his
name
right
now,
I'm
like
I,
am
embarrassed
but
I'm
also
terrible
with
names
so
but
yeah,
so
he
he.
He
was
like
in
his
college
campus.
He
was
in
like
the
I.T
room
or
whatever,
and
he
would
help
people
out.
People
be
like.
B
Oh
go
look
for
the
guy
in
the
red
hat
because
he
wore
a
red
baseball
cap,
and
so,
as
the
company
evolved,
it
became
or
well
as
he
decided
to
create
a
company
that
was
where
he
got
the
naming
from
I.
Think
I
literally
have
this
story
and
like
my
story
of
Red
Hat
book
that
should
exist
on
my
desk,
that
they
ship
you
with
your
Fedora
but
anyway.
B
A
B
A
I
love
it.
For
that
reason,
I
mean
we
try
and
think
about
like
also
what
we're
going
to
do
at
conferences,
and
we
want
to
do
usually
something.
That's
like
a
little
educational
like
the
T-shirt,
but
like
also
spreads
awareness
about
Robusta,
so
I
I,
don't
think
we're
ever
going
to
be
able
to
find
something,
though,
as
good
as
the
red
hat.
B
That
was
that
was
that
was
that
was
a
good
stroke,
but
it's
an
interesting
story
behind
why
red
hat
is
a
red
hat
I,
don't
remember
why
the
why
the
change
of
the
Fedora,
but
they
also
so
at
the
very
beginning
of
red
hat.
They
sold
a
lot
of
t-shirts.
It
was
Mark
Ewing's,
hat
okay.
Thank
you.
I
appreciate
the
smarter
people
than
I,
with
better
memories
or
better
faster
Google
skills,
whichever
it
was
so
yeah,
so
the
the
we
did
actually
sell
a
lot
of
t-shirt
things.
B
A
B
A
A
Yeah
there's
some
trick
for
there,
though
oh
a
game
where
you
have
to
get
between
two
different
articles
and
the
shortest
path
possible,
and
then
the
trick
is
like
the
trick
is
to
always
like
aim
to
get
to
philosophy
or
something
there's
some
like
trick
there
for,
like,
like
a
mega,
join
your
page
that
connects
everything.
Okay,.
B
A
I
want
to
throw
off
just
one
more
thing
before
we
stop
and
I
want
to
speak
a
little
bit
more
about
something
else.
That
happens
all
the
time.
You
have
an
issue
that
occurs
in
one
cluster
and
doesn't
occur
in
another
cluster,
and
you
say
why
is
it
different
about
this
cluster
from
all
the
other
clusters
ways
different
this
sorry?
What
is
different
about
this
cluster,
so
I
mean
as
part
of
our
vision
here
for
giving
you
like
this
One-Stop
troubleshooting
platform.
A
Where
you
go
and
you
have
any
issue
in
production,
then
part
of
that
is
actually
we
added
on
the
feature.
Recently,
we
let
you
compare
different
clusters,
so
I'll
just
show
you
what
that
looks
like
so
I
can
take
care.
I
can
either
compare
different,
namespaces
or
different
clusters,
and
that
really
also
corresponds
to
some
two
different
ways
that
people
sometimes
do
multi-tenancy.
So
I'm
gonna,
assume
I,
have
two
clusters
I'm
just
going
to
select
here.
Two
clusters
and
I
can
see
here
comparison.
Let's
hide
everything.
A
That's
identical,
so
I
can
see
here,
like
the
exact
difference.
I
can
see
here,
for
example,
in
this
cluster
I'm
running
an
older,
robust
Runner
version
and
here's
the
image
on
the
new
one,
so
it
makes
it
really
easy
to
just
pinpoint
in
a
second
what
the
exact
differences
are
when
you
have
another,
that's
fire
ink.
B
B
The
actual
production
issue
we
had
three
weeks
ago:
I
wasn't
the
person
who
actually
handled
that
issue
myself
so
I,
don't
remember
what
the
INS
What,
what
the
like
subtle
difference
ended
up
being
that
caused
the
problem,
but
yeah
tools,
tools
to
sit
there
and,
like
literally,
compare
okay,
we've
got
this
this
customer.
They
have
these
two
clusters.
Their
clusters
are
basically
identical.
So
where
is
where?
Is
that
super
secret
difference?
That's
causing
the
problem
and
that
that's
that's.
B
B
A
Right
most
of
the
time
you
are
more
experienced
than
me,
especially
when
it
comes
to
SRE
stuff,
so
I
feel
like
an
imposter,
sometimes
because
I
didn't
have
that
much
of
a
like
I
have
a
lot
of
software
engineering
background,
but
somehow
I,
like
jump
from
there
to
then
ended
up
in
a
cult
and
a
co-founder
role
like
without
doing
as
many
years
of
SRE
as
you
so
I
mean
I,
I.
Think
you're,
probably
right
on
most
of
the
issues
that
we
end
up
disagreeing
about.
B
And
I
think
when
we're,
when
we
disagree,
sometimes
there's
places
that
reasonable
people
can
disagree,
but
either
way
we
have
a.
We
have
a
great
deal
of
fun,
arguing
with
each
other
on
the
internet.
I
have
an
immense
amount
of
respect
and
I
like
respect
for
you
and
I.
Like
you
a
lot
so
we'll
have
to
have
you
back
sometime.
B
All
right,
thank
you.
Folks
and
we'll
see
you
in
two
weeks.
Oh
Christian
Hernan
wait
is
Christian
Hernandez
joining
us.
Christian
Hernandez
is
joining
us
on
August
11th
to
talk
about
his
book
about
get
UPS.
What
was
the
book
The
entitled
book?
I
am
the
worst
I'll
post
it
on
Twitter
and
be
a
much
better
host
for
you
all
after
my
vacation,
I'm
already
vacation
brained.
That's
it.
A
Okay,
so
I
don't
know
if
you
can
yeah
with
the
screenshot
I'm,
actually
taking
a
green
shot
of
the
screenshot
I'm,
taking
a
screenshot
of
this
to
send
to
my
product
guy
it.
The
answer
is
not
yet,
but
you
can
read
between
the
lines
on
that
and
if
it's
something
you're
interested
in
then
please
please.
Please
message
me
on
LinkedIn
or
on
Twitter
or
reach
out,
or
you
have
a
slot
Community
as
well,
so
come
and
talk
to
us
there,
because
we
have
a
product
guy.